Framework & Assessment

The AI Context
Maturity Framework

Where is your enterprise AI programme — and what's stopping it reaching production? Five questions. Find your stage.

Take the assessment → Read the framework
The State of Enterprise AI

It is not the model.
It is not the data.

Enterprise AI is failing at scale. Not because the models are wrong. Because the context is missing — the structured business meaning that tells AI what data represents, what rules apply, and who is accountable.

BCBS 239
risk data aggregation requires accurate, complete, timely, and adaptable information
Basel Committee
SR 11-7
model risk frameworks require documentation, validation, controls, and clear limitations
Federal Reserve
EU AI Act
high-risk AI systems require technical documentation, oversight, and lifecycle controls
European Commission

The pipeline era had a hidden advantage: the pipeline itself was a translation layer. AI does not use the pipeline. AI goes directly to the data. And when it does, context is the only thing that stands between a confident answer and a correct one.

Without context, AI guesses. With it, AI operates.

The Starting Point

The investment is real.
It has not been activated.

Most financial services firms have already done the work. Years of metadata governance: business terms defined, lineage traced, ownership assigned, rules documented. Built for compliance. Never designed to be consumed by machines.

Documented metadata
Sits in a catalogue
Consulted on a review cycle
A reference — you go, you look, you leave
Maintained by the governance team
Built for compliance reporting
Operational metadata
Travels with the data
Present at point of decision
The material AI works with, at runtime
Maintained continuously as business changes
Built for machine consumption

The governance investment is not wasted. The question is whether it has been activated — converted from a reference system into something a machine can act on.

This transition — from governed metadata to operational context — sits at the heart of what the market is beginning to call AI enablement: the emerging layer of the data intelligence stack that connects governance investment to AI performance.
The 5-Stage Framework

Where does your programme sit?

The gap between where most enterprises are and where their AI programmes need them to be can be mapped precisely. Click each stage to explore it.

Define

Business terms, classifications, and glossary established. Data ownership assigned. Governance policies documented and reviewed on a cycle. This is where most enterprises begin — and where the majority of governance investment is concentrated.

Stage 1
What you have

A data catalogue. Defined business terms. Ownership accountability. Policies reviewed annually. Compliance evidence produced on request.

Where programmes stall

Meaning is captured but not connected. The catalogue is a reference, not an operating system. Context exists for humans — not yet for machines.

Enrich

Relationships, lineage, ownership, and transformation rules connected across the data estate. Data products emerging. The metadata graph is active — a structured map of how data flows and what governs it.

Stage 2
What you have

An active metadata graph. Traced lineage. Ownership mapped to data domains. Business rules linked to the data they govern.

Where programmes stall

The graph reflects how IT describes data. Not how the business uses it. Metadata is richer — but still documented, not operational.

Embed — where the supply chain breaks

This is where the context supply chain breaks. The task at Stage 3 is translating governance — defined terms, traced lineage, documented rules — into machine-readable context that AI can consume directly.

Stage 3
What you have

Partial. Some context embedded in pipelines or transformation logic, but inconsistently. AI systems access the data estate — but the context governing what that data means does not travel with it.

Why most programmes stall here

The context exists in the governance layer. It does not travel to where AI operates. This is an architectural gap, not a data quality problem. Most enterprise AI programmes are operating here.

Serve

Context delivered at runtime to AI systems and applications. When an AI agent reaches the data, it arrives with the full context: meaning, provenance, rules, ownership, and currency — live, not in a catalogue.

Stage 4
What you have

AI systems that operate within governed constraints. Auditability that is automatic, not reconstructed after the fact. The governance investment finally consumed by the systems it was designed to govern.

What prevents reaching Stage 4

Serving context at runtime requires architectural decisions most enterprises have not yet made: how context is structured, versioned, and exposed to AI at the point of consumption.

Maintain

Context updates continuously as business processes change. When a regulatory rule is revised or a new data product introduced — context updates automatically. Governance is a byproduct of the workflow, not layered on top of it.

Stage 5
What you have

An AI programme that stays current without manual intervention. Compliance that is continuous, not periodic. Governance that nobody notices — because it is embedded, not enforced.

"The best governance is the kind nobody notices."
Self-Assessment

Find your stage

Answer five questions honestly. Your score maps directly to the framework — and gives you a clear picture of where your AI programme stands today.

Score each question:

1 = Not in place    2 = Partially    3 = Fully operational
1Can you define — in machine-readable terms — what any data element means in the context of a specific business process?
2Is data lineage connected to business ownership and transformation rules — not just technical metadata?
3Does business context travel with your data at the point AI or an application consumes it?
4Can an AI system access the meaning, rules, and accountability for the data it's working with — at runtime?
5When business processes change, does your context update automatically — without a manual governance cycle?
0 of 5
Your result

Let's talk about what comes next

If this framework is useful, I'd welcome a conversation. I work with data and AI leaders in financial services on exactly this problem.

Connect with Stuart on LinkedIn
Read more xflow insight →